letters to nature

Methods 23. Nørskov, J. K. et al. Universality in heterogeneous catalysis. J. Catal. 209, 275–278 (2002). 24. Da Silva, J. L. F., Stampfl, C. & Scheffler, M. Adsorption of Xe atoms on metal surfaces: new insights The growth experiments were carried out in a Philips CM300 FEG TEM equipped with an from first-principles calculations. Phys. Rev. Lett. 90, 066104 (2003). in situ cell13. This system provides images with a resolution better than 0.14 nm of samples 25. Vanderbilt, D. H. Soft self-consistent pseudopotentials in a generalized eigenvalue formalism. Phys. exposed to reactive gases and elevated temperatures. ATietz FastScan-F114 CCD was used Rev. B 41, 7892–7895 (1990). for recording TEM images. The image acquisition time was adjusted to suit the reshaping 21 26. Kresse, G. & Forthmu¨ller, J. Efficiency of ab-initio total energy calculations for metals and of the Ni nanocrystals and ranged from 2 to 10 frames s . A total of 16 video sequences semiconductors using a plane-wave basis set. Comput. Mater. Sci. 6, 15–50 (1996). (each consisting of 300–1,200 consecutive high-resolution images) showing the coupling of graphene growth to Ni step-edge dynamics were recorded in six different growth experiments. The electron beam intensity was kept in the interval 0.2–2 A cm22, which was Supplementary Information accompanies the paper on www.nature.com/nature. sufficiently low to avoid any influence of the electron beam on the growth. Growth experiments with the electron beam switched off resulted in the same carbon nanofibre Acknowledgements We gratefully acknowledge the participation of the CTCI Foundation, structures as in the low-dose in situ experiments. Taiwan, in the establishment of the TEM facility at Haldor Topsøe A/S. S.H. acknowledges The self-consistent DFT calculations were performed using the LDA14 or the GGA- support from the Danish Research Council, STVF. F.A.P. and J.K.N. acknowledge computational RPBE approximation16 for exchange and correlation. The interaction of the graphene resources from the Danish Center for Scientific Computing. C.L.-C. acknowledges support from overlayer with the Ni(111) surface was dominated by van der Waals forces. The LDA for Facultad de Ciencias, Universidad de Ca´diz, Puerto Real, Spain, for a sabbatical visit to Haldor the exchange and correlation term is known to give the best description of such weak Topsøe A/S. interactions24. We have therefore used LDA for calculating the adsorption energy and diffusion energy barrier for C and Ni adatoms on the clean Ni(111) surface and the Competing interests statement The authors declare competing financial interests: details graphene–Ni(111) interface. The adsorption energies and diffusion energy barriers on the accompany the paper on www.nature.com/nature. clean Ni(111) surface calculated by the GGA-RPBE approximation were found to be essentially the same. However, within the GGA-RPBE approximation we found no Correspondence and requests for materials should be addressed to S.H. ([email protected]). adhesion of the graphene overlayer to the Ni(111) surface. The ionic cores are described by ultrasoft pseudopotentials25 and the one-electron valence states were expanded in a basis of plane waves with kinetic energy below 25 Ry. The electron density was determined self-consistently by iterative diagonalization of the Kohn–Sham hamiltonian, Pulay mixing26 of the resulting electronic density and Fermi occupation of the Kohn–Sham states (k BT ¼ 0.1 eV). All total energy calculations were ...... extrapolated to zero electronic temperature and performed with the magnetic moment of the nickel surface taken into account. Enhanced ice sheet growth in Eurasia The Ni(111) surface was modelled by a periodic repetition of a three-layer slab, whereas a slab thickness of nine layers was used in the [211] direction to create a stepped surface and a thickness comparable to the Ni(111) surface. In both cases, the slabs were separated owing to adjacent ice-dammed lakes by ,10 A˚ of vacuum. In the Ni(111) model, the top layer was allowed to fully relax, whereas in the Ni(211) model, the uppermost close-packed (111) layer was relaxed fully. G. Krinner1, J. Mangerud2, M. Jakobsson3, M. Crucifix4*, C. Ritz1 The remaining layers were kept fixed in the bulk geometry with a calculated equilibrium & J. I. Svendsen2 lattice constant of 3.52 A˚ . For all calculations on Ni(111), a p(2 £ 2) lateral unit cell was used, giving C and Ni adsorbate coverages of 0.25 ML, whereas the C/Ni ratio for the 1LGGE, CNRS-UJF Grenoble, BP 96, F-38402 Saint Martin d’He`res, France graphene overlayer was 2. For the Ni(211) surface, a (1 £ 2) unit cell was used in the 2 calculations with a one-to-one coverage of adsorbate to Ni step edge atoms. A sampling of Department of Earth Science, University of Bergen, and Bjerknes Centre for 4 £ 4 £ 1 Monkhorst–Pack special k-points was used, resulting in 8 k-points in the Climate Research, Alle´gt. 41, N-5007 Bergen, Norway 3 irreducible Brillouin zone. Center for Coastal and Ocean Mapping/Joint Hydrographic Center, Chase Ocean Engineering Laboratory, 24 Colovos Road, Durham, New Hampshire 03824, Received 20 August; accepted 5 December 2003; doi:10.1038/nature02278. USA 1. Baughman, R. H., Zakhidov, A. A. & de Heer, W. A. Carbon nanotubes—the route towards 4Institut d’Astronomie et de Ge´ophysique G. Lemaıˆtre, Universite´ catholique de applications. Science 297, 787–792 (2002). Louvain, chemin du Cyclotron, 2, B-1348 Louvain-la-Neuve, Belgium 2. Ajayan, P. M. Nanotubes from carbon. Chem. Rev. 99, 1787–1799 (1999).

3. Rostrup-Nielsen, J. R., Sehested, J. & Nørskov, J. K. Hydrogen and synthesis gas by steam and CO2 * Present address: Met Office, Hadley Centre for Climate Prediction and Research, Fitzroy Road, Exeter reforming. Adv. Catal. 47, 65–138 (2002). EX1 3PB, UK 4. Ebbesen, T. W. & Ajayan, P. M. Large-scale synthesis of carbon nanotubes. Nature 358, 220–222 ...... (1992). Large proglacial lakes cool regional summer climate because of 5. Thess, A. et al. Crystalline ropes of metallic carbon nanotubes. Science 273, 483–487 (1996). their large heat capacity, and have been shown to modify 6. Kong, J., Soh, H. T., Cassell, A. M., Quate, C. F. & Dai, H. Synthesis of individual single-walled carbon nanotubes on patterned silicon wafers. Nature 395, 878–881 (1998). precipitation through mesoscale atmospheric feedbacks, as in 1 7. Boskovic, B. O. et al. Large-area synthesis of carbon nanofibres at room temperature. Nature Mater. 1, the case of . Several large ice-dammed lakes, with a 165–168 (2002). combined area twice that of the Caspian Sea, were formed in 8. Baker, R. T. K., Harris, P. S., Thomas, R. B. & Waite, R. J. Nucleation and growth of carbon deposits northern Eurasia about 90,000 years ago, during the last glacial from the nickel catalyzed decomposition of acetylene. J. Catal. 26, 51–62 (1972). 2 9. Rostrup-Nielsen, J. R. Equilibria of decomposition reactions of carbon monoxide and methane over period when an ice sheet centred over the Barents and Kara seas nickel catalysts. J. Catal. 27, 343–356 (1972). blocked the large northbound Russian rivers3. Here we present 10. Charlier, J.-C. & Iijima, S. in Growth Mechanisms of Carbon Nanotubes (eds Dresselhaus, M. S., high-resolution simulations with an atmospheric general circu- Dresselhaus, G. & Avouris, Ph.) Vol. 80, Topics in Applied Physics 55–80 (Springer, Heidelberg, 2001). 11. Clausen, B. S., Topsøe, H. & Frahm, R. Application of combined X-ray diffraction and absorption lation model that explicitly simulates the surface mass balance of techniques for in situ catalyst characterisation. Adv. Catal. 42, 315–344 (1998). the ice sheet. We show that the main influence of the Eurasian 12. Topsøe, H. In situ characterisation of catalysts. Stud. Surf. Sci. Catal. 130, 1–21 (2000). proglacial lakes was a significant reduction of ice sheet melting at 13. Hansen, P. L. et al. Atom-resolved imaging of dynamic shape changes in supported copper the southern margin of the Barents–Kara ice sheet through nanocrystals. Science 295, 2053–2055 (2002). 14. Jones, R. O. & Gunnarsson, O. The density functional formalism, its applications and prospects. Rev. strong regional summer cooling over large parts of Russia. In Mod. Phys. 61, 689–746 (1989). our simulations, the summer melt reduction clearly outweighs 15. Rosei, R. et al. Structure of graphitic carbon on Ni(111): A surface extended-energy-loss fine-structure lake-induced decreases in moisture and hence snowfall, such as study. Phys. Rev. B 28, 1161–1164 (1983). has been reported earlier for Lake Agassiz1. We conclude that the 16. Hammer, B., Hansen, L. B. & Nørskov, J. K. Improved adsorption energetics within density-functional theory using revised Perdew-Burke-Ernzerhof functionals. Phys. Rev. B 59, 7413–7421 (1999). summer cooling mechanism from proglacial lakes accelerated ice 17. Bengaard, H. S. et al. Steam reforming and graphite formation in Ni catalysts. J. Catal. 209, 365–384 sheet growth and delayed ice sheet decay in Eurasia and probably (2002). also in North America. 18. Trimm, D. L. The formation and removal of coke from nickel catalyst. Catal. Rev. Sci. Eng. 16, 155–189 We used the LMDZ3 (Laboratoire de Me´te´orologie Dynamique, (1977). 19. Alstrup, I. A new model explaining carbon filament growth on nickel, iron and Ni-Cu alloy catalysts. CNRS Paris) stretched-grid atmosphere general circulation model J. Catal. 109, 241–251 (1988). (AGCM). In addition to a present-day reference simulation (called 20. Rostrup-Nielsen, J. R. & Trimm, D. L. Mechanisms of carbon formation on nickel-containing ‘PD’), two simulations for 90 kyr before present (BP) were carried catalysts. J. Catal. 48, 155–165 (1977). out. Both were constrained by appropriate ice sheet reconstructions, 21. De Jong, K. P. & Geus, J. W. Carbon nanofibers: catalytic synthesis and applications. Catal. Rev. Sci. Eng. 42, 481–510 (2000). sea surface conditions, greenhouse gas concentrations and insola- 22. Askeland, D. R. The Science and Engineering of Materials 2nd edn, Ch. 5 (PWS-KENT, Boston, 1989). tion as boundary conditions (Fig. 1; see Methods for details). They 429 NATURE | VOL 427 | 29 JANUARY 2004 | www.nature.com/nature © 2004 Nature Publishing Group letters to nature differ by the fact that experiment ‘L’ (for ‘lakes’) includes Russian margin decreases the horizontal air density gradient, and therefore ice-dammed lakes3 and the Baltic lake4, while these are replaced by induces a weakening of the katabatic circulation in summer. Above normal continental surface in simulation ‘NL’ (for ‘no lakes’). The the lakes, the summer cooling is associated with a substantial surface grid-stretching capability of LMDZ3 allows high resolution pressure increase (þ4 mbar). The summer cooling, and, to a lesser (100 km) over the Barents–Kara region. Compared with observed degree, the reduced surface wind speed on the ice sheet slope present-day surface air temperature5 and precipitation6, the mean (leading to lower downward turbulent surface heat fluxes), strongly absolute annual temperature error in simulation PD over the region reduce summer melt along the southern margin of the Barents–Kara of interest is fairly low (1.0 8C), but annual precipitation is over- ice sheet and the southeastern margin of the Scandinavian ice sheet. estimated by about 60%, mainly because of a wet bias in winter. On On the lakeward parts of the ice sheet, runoff is reduced by the other hand, LMDZ3 quite adequately captures the present 8mmd21 in July, and by more than 500 mm yr21 (50%) in the surface mass balance patterns and quantities over the present ice annual mean, when the lakes are taken into account (refreezing of sheets7,8. melt water is accounted for in calculating runoff from snow and ice Analysis of the upper soil temperatures in the region in simu- melt12). Sensitivity tests showed that both the sign and the absolute lation L, based on a normalized surface frost index9, shows that the value of the annual mean precipitation change are dependent on the Russian proglacial lakes were mostly situated within the limit of prescribed sea-surface boundary conditions, but the precipitation continuous permafrost, only the southernmost part of the West change always remains far smaller than the ablation decrease. The Siberian Lake (Fig. 1) extending into a region with extensive latter coherently dominates the simulated surface mass balance discontinuous permafrost. This is in agreement with ice wedge change (Fig. 2b). casts found on palaeo-beaches10. In both L and NL, simulated The Barents–Kara and the Scandinavian ice sheets were linked by precipitation 90 kyr ago is less than 50% of PD over large parts a relatively small ice bridge at about 35 8E. Their surface mass of Russia. Simulated inflow of ice sheet melt water is substantial in balances (SMB, Fig. 3) are therefore analysed separately. The L, corresponding to more than 1 m yr21 lake level rise. This is integrated SMB of the Barents–Kara ice sheet, defined as precipi- supported by findings of ice marginal deltas formed in the lakes11. tation minus evaporation/sublimation and minus runoff, is 21 Simulated lake summer surface temperatures exceed 4 8C only near þ0.18 m of water equivalent per year (mw yr )inL,and 21 the southern shore. In their northern parts, the proglacial lakes are 20.43 mw yr in NL. According to a paired Student’s t-test with cold monomictic (one vertical mixing per year, in summer; bottom nine degrees of freedom, the confidence in the SMB increase in L temperature below 4 8C). They remain ice-covered from 7 months with respect to NL exceeds 99%, which excludes our result from per year on the southern shore to 11 months per year near the ice being an artefact of interannual variability. Furthermore, the sheet in spite of high meltwater input rates. The Baltic Lake, southern ablation zone of the Barents–Kara ice sheet extends over situated at lower latitudes, is somewhat warmer, and the ice cover lasts between 6 months on the southern shore and 9 months near the ice sheet. During all seasons, surface winds over the ice sheet in L and NL show a typical katabatic pattern, the air masses flowing down the slope of the ice sheet, deviated to the right because of the Earth’s rotation, with maximum speeds near the ice sheet margin. The primary atmospheric effect of the lakes is a strong summer cooling due to their large heat capacity. Summer surface air temperature anomalies (L–NL) exceed 210 8C above the lakes. The impact over the southern margin of the Barents–Kara ice sheet is substantial up to an altitude of about 2,000 m (Fig. 2a). This cooling effect extends horizontally about 1,000 km away from the lakes. The accumulation of cool air off and above the ice sheet

Figure 2 Lake-induced climate anomalies between simulation L and simulation NL. a, JJA (June, July, August) air temperature anomalies: vertical cut through the lower atmosphere at 708 E. The pattern is representative for all longitudes where proglacial lakes are adjacent to the Barents–Kara ice sheet. b, Monthly mean anomalies of surface Figure 1 Maximum ice sheet and lake extent in northern Russia during the early mass balance (SMB) components precipitation (black), runoff (red) and evaporation/ Weichselian, about 90,000 yr ago. Abbreviations denote the West Siberian Lake (WS), sublimation (green) on the Scandinavian/Barents–Kara ice sheet grid points that belong to Lake Komi (LK), the White Sea Basin Lake (WSB), and the Baltic Lake (B). the lakes’ drainage basin.

430 © 2004 Nature Publishing Group NATURE | VOL 427 | 29 JANUARY 2004 | www.nature.com/nature letters to nature several grid points towards the interior in simulation NL, and decreased surface mass balance of the . However, 21 significant melt (in excess of 0.2 mw yr ) occurs in both simu- assessing the surface mass balance precisely was not possible in that lations at least two grid points from the ice limit. Hence, the study as ice melt was not explicitly calculated (S. W. Hostetler, resolution is sufficient to capture melt processes, such that the personal communication). Because the same study suggested that sensitivity of the SMB can be better trusted than in most AGCMs Lake Agassiz induced a significant regional cooling in summer, that hardly represent the ablation zone. Concerning the Scandina- similar to the cooling induced by the Russian lakes reported here, we vian ice sheet, the Russian proglacial lakes and the Baltic Lake postulate that the lake-induced reduction in melting on the 21 southern margin of the Laurentide ice sheet was larger than the induce a substantial increase (þ1.7 mw yr ) of the simulated SMB owing to reduced melt rates. Note that the prescribed western reduction of precipitation. Furthermore, we see no reason why the margin of the Scandinavian ice sheet does not reach the Atlantic feed-back mechanism reported here should not have operated coastline, and that outlet glaciers in fjords and valleys are not during earlier growth and decay phases of the Laurentide ice sheet, during which large ice-dammed lakes are known to have prescribed, owing to insufficient resolution. The westernmost 13 ice sheet grid points are therefore at altitudes (more than existed . The main conclusion of this study is that the Russian proglacial 1,000 m, mostly more than 1,500 m) where significant ablation lakes, through their effect on the surface mass balance of the does not occur, which explains the high surface mass balance Barents–Kara ice sheet, repeatedly played an important role in the obtained here. High melt rates and therefore negative surface regional climate dynamics. Inception of the Barents–Kara ice mass balances are simulated on a few low-altitude grid points sheet was probably caused by insolation changes and positive near Franz Josef Land. Given the high latitude, the great distance feedbacks involving snow albedo, ocean circulation14,veg- to the Atlantic, the vicinity of the glacial Arctic Ocean, the ice etation15,16 and greenhouse gas concentration changes17. As soon sheet to the south, and low CO2 concentrations, this result seems as the ice sheet was big enough to dam the northward flowing unlikely. Russian rivers, thereby creating the proglacial lakes, this initial Although it focused on a different climatic and regional context, growth of the Barents–Kara ice sheet was notably enhanced. it is interesting to note that a study of the North American climate at Optical stimulated luminescence dates indicate that the ice 11,000 yr BP (ref. 1) suggested that the presence of sheet attained its maximum extent between 90 and 80 kyr Agassiz caused a reduction of precipitation rates on the adjacent ago2,10. On the other hand, June insolation at 65 8N, which is Laurentide ice sheet (the magnitude of which is similar to the close to the southern limit of the Barents–Kara ice sheet, precipitation changes induced by the Russian proglacial lakes, and increased by ,12% between 95 and 85 kyr ago18. The most much weaker than the ablation reduction reported here). The coherent scenario is therefore that the ice sheet attained its maxi- authors concluded that this precipitation reduction led to a mum extent at about 90 kyr ago, and that a subsequent retreat (delayed, but not stopped by the lake-induced melt reduction) began around 85 kyr ago owing to increased summer insolation. Once the Barents–Kara ice sheet had shrunk sufficiently, rapid subglacial drainage of the lakes occurred. This further accelerated the deglaciation by leading to higher summer melt rates along the southern ice sheet margin. Note that rapid drainage of proglacial lakes along the southern margin of the Laurentide ice sheet is thought to have caused abrupt widespread cooling by interrupting the oceanic thermohaline cir- culation19; it has been suggested that a similar cooling, possibly compensating for the regional warming caused by the drainage of the lakes, might have occurred following lake drainage in northern Eurasia11. During later stages of the last glacial, in particular the Last Glacial Maximum (about 20,000 yr ago), the Barents–Kara ice sheet did not grow big enough to create large proglacial lakes11 (possibly because the larger Scandinavian ice sheet blocked moisture delivery from the Atlantic), and the growth acceleration due to these lakes therefore did not take place. Finally, we note that the coupled lake– ice sheet system could be self-sustaining to a certain degree: a small ice sheet decay leads to an increased lake size, thereby cooling the climate and thus reducing the meltwater production. This negative feedback will break down if the ice sheet decay is sufficiently pronounced to allow drainage of the lake through previously blocked passages. A Methods Model The LMDZ3 model has been run with 96 £ 72 (longitude £ latitude) grid points. The chosen nominal horizontal resolution is irregular, varying from 100 km at the centre of the region of interest (between 08 and 120 E and between 558 and 858 N) to about 550 km roughly at the antipodes of this region. The model has 19 vertical layers in the atmosphere. LMDZ3 has been adapted for the simulation of cold climates20 and includes a thermal lake module21.

Figure 3 Simulated ice sheet surface mass balance and lake-induced surface mass Boundary conditions The prescribed boundary conditions (Fig. 1) for the AGCM include orbital parameters for balance anomaly. a, Simulated surface mass balance in simulation L; b, lake-induced 18 90 kyr ago ; an atmospheric CO2 concentration of 205 p.p.m.v. (ref. 22); reconstructed anomaly (L–NL). Units are metres of water equivalent per year. Bold lines, ice sheets and dimensions of the Barents–Kara and the Scandinavian ice sheet based on geological lakes (grid-scale fraction . 20%) for 90 kyr ago; thin line, present-day coastline. evidence2 and glaciological modelling23 with some modifications to fit the latest results

431 NATURE | VOL 427 | 29 JANUARY 2004 | www.nature.com/nature © 2004 Nature Publishing Group letters to nature obtained within the ESF-QUEEN project; ice sheet thicknesses for the rest of the Northern ...... Hemisphere obtained using the climatic index method24 applied to an ice sheet model and AGCM outputs25; modelled global sea surface temperatures and sea-ice extent26, surface Natural examples of olivine lattice albedo and roughness over ice-free surfaces derived from modelled glacial maximum palaeovegetation27; a large lake in the part of the Baltic Sea depression not covered by the Scandinavian ice sheet4; and reconstructed proglacial lakes in European Russia and in West preferred orientation patterns with Siberia3. a flow-normal a-axis maximum Initial conditions and spin-up The simulations were started from an initial state with present-day soil temperatures. Tomoyuki Mizukami1,2, Simon R. Wallis2 & Junji Yamamoto3* From the 15 yr of each simulation, the first five years were discarded as spin-up. Lake temperatures at the bottom level (between 15 and 30 m) were in equilibrium at the end of 1Department of Geology and Mineralogy, University of Kyoto, Sakyo-ku, Kyoto, the 5 spin-up years. At the end of each of the first three spin-up years for the 90-kyr-ago 606-8501, Japan simulations, soil temperature, being the slowest simulated component of the climate 2Department of Earth and Planetary Sciences, Graduate School of Environmental system simulated here, was separately spun up for 1,000 yr using the year’s modelled monthly surface temperatures. Studies, University of Nagoya, Chikusa-ku, Nagoya, 464-8601, Japan 3Laboratory of Earthquake Chemistry, University of Tokyo, Bunkyo-ku, Tokyo, Received 15 May; accepted 14 November 2003; doi:10.1038/nature02233. 113-0033, Japan 1. Hostetler, S. W., Bartlein, P. J., Clark, P. U., Small, E. E. & Solomon, A. M. Simulated influences of * Present address: Department of Earth and Planetary Sciences, Tokyo Institute of Technology, Lake Agassiz on the climate of central North America 11,000 years ago. Nature 405, 334–337 (2000). Meguro-ku, Tokyo, 152-8551, Japan 2. Svendsen, J. I. et al. Late Quaternary ice sheet history of Eurasia. Quat. Sci. Rev (in the press)...... 3. Mangerud, J., Astakhov, V., Jakobsson, M. & Svendsen, J. I. Huge ice-age lakes in Russia. J. Quat. Sci. 16, 773–777 (2001). Tectonic plate motion is thought to cause solid-state plastic flow 4. Lundquist, J. Glacial stratigraphy in Sweden. Spec. Pap. Geol. Surv. Finl. 15, 43–59 (1992). within the underlying upper mantle and accordingly lead to the 5. Legates, D. R. & Willmott, C. J. Mean seasonal and spatial variability in global surface air temperature. development of a lattice preferred orientation of the constituent Theor. Appl. Climatol. 41, 11–21 (1990). 1–3 6. Legates, D. R. & Willmott, C. J. Mean seasonal and spatial variability in gauge-corrected, global olivine crystals . The mechanical anisotropy that results from precipitation. Int. J. Climatol. 10, 111–127 (1990). such preferred orientation typically produces a direction of 7. Genthon, C. & Krinner, G. Antarctic surface mass balance and systematic biases in general circulation maximum seismic wave velocity parallel to the plate motion models. J. Geophys. Res. 106, 20653–20664 (2001). direction4,5. This has been explained by the existence of an olivine 8. Krinner, G. & Werner, M. Impact of precipitation seasonality changes on isotopic signals in polar ice cores: A multi-model analysis. Earth Planet. Sci. Lett. 216, 525–538 (2003). preferred orientation with an ‘a-axis’ maximum parallel to the 9. Nelson, F. & Outcalt, S. I. A computational method for prediction and regionalization of permafrost. induced mantle flow direction3,5,6–8. In subduction zones, how- Arct. Alp. Res. 19, 279–288 (1987). ever, the olivine a axes have been inferred to be arranged roughly 10. Mangerud, J., Astakhov, V., Murray, A. & Svendsen, J. I. The chronology of a large ice-dammed 9–13 lake and the Barents–Kara ice sheet advances, Northern Russia. Glob. Planet. Change 31, 321–336 perpendicular to plate motion , which has usually been 10–13 (2001). ascribed to localized complex mantle flow patterns . Recent 11. Mangerud, J. et al. Ice-dammed lakes and rerouting of the drainage of northern Eurasia during the last experimental work14 suggests an alternative explanation: under glaciation. Quat. Sci. Rev. (in the press). conditions of high water activity, a ‘B-type’ olivine preferred 12. Thompson, S. L. & Pollard, D. Greenland and Antarctic mass balances for present and doubled

atmospheric CO2 from the GENESIS version-2 global climate model. J. Clim. 10, 871–900 (1997). orientation may form, with the a-axis maximum perpendicular 13. Teller, J. T. in The Quaternary Period in the United States (eds Gillespie, A., Porter, S. & Atwater, B.) to the flow direction. Natural examples of such B-type preferred Ch. 3 (Elsevier, Amsterdam, 2003). orientation are, however, almost entirely unknown. Here we 14. Khodri, M. et al. Simulating the amplification of orbital forcing by ocean feedbacks in the last glaciation. Nature 410, 570–574 (2001). document widespread B-type olivine preferred orientation pat- 15. de Noblet, N. et al. Possible role of atmosphere-biosphere interactions in triggering the last glaciation. terns from a subduction-type metamorphic belt in southwest Geophys. Res. Lett. 23, 3191–3194 (1996). Japan and show that these patterns developed in the presence of 16. Gallimore, R. G. & Kutzbach, J. E. Role of orbitally induced changes in tundra area in the onset of water. Our discovery implies that mantle flow above subduction glaciation. Nature 381, 503–505 (1996). 17. Shackleton, N. J. The 100,000-year ice-age cycle identified and found to lag temperature, carbon zones may be much simpler than has generally been thought. dioxide, and orbital eccentricity. Science 289, 1897–1902 (2000). The Higashi-akaishi peridotite body of southwest Japan is the 18. Berger, A. Long-term variations of daily insolation and Quaternary climatic changes. J. Atmos. Sci. 35, only kilometre-scale garnet peridotite body yet found in a subduc- 2362–2367 (1978). 15 19. Clarke, G., Leverington, D., Teller, J. & Dyke, A. Superlakes, megafloods, and abrupt climate change. tion-type metamorphic belt . This body offers an unrivalled Science 301, 922–923 (2003). opportunity to study subduction zone mantle directly. Here we 20. Krinner, G., Genthon, C., Li, L. & Le Van, P. Studies of the Antarctic climate using a stretched-grid focus on the olivine lattice preferred orientation (LPO) patterns of general circulation model. J. Geophys. Res. 102, 13731–13745 (1997). this body. The Higashi-akaishi body is tabular with an area of 21. Krinner, G. Impact of lakes and wetlands on boreal climate. J. Geophys. Res. D 108, 101029/ 2 2002JD002597 (2003). 10 km and a thickness of 500 m (ref. 16) and occurs as an integral 22. Petit, J.-R. et al. Climate and atmospheric history of the past 420,000 years from the Vostok ice core, part of the Cretaceous Sanbagawa high-pressure, low-temperature Antarctica. Nature 399, 426–436 (1999). metamorphic belt17–19. Dunite, a rock type of more than 90% 23. Siegert, M. J., Dowdeswell, J. A., Hald, M. & Svendsen, J. I. Modelling the Eurasian ice sheet through a full (Weichselian) glacial cycle. Glob. Planet. Change 31, 367–385 (2001). olivine, is the main constituent of the body with minor amounts 24. Marshall, S. J., Tarasov, L., Clarke, G. & Peltier, W. R. Glaciological reconstruction of the Laurentide of wehrlite, pyroxenite, garnet pyroxenite, chromitite and garnet Ice Sheet: Physical processes and modelling challenges. Can. J. Earth Sci. 37, 769–793 (2000). peridotite15,16,20. Serpentinized equivalents of these rock types devel- 25. Charbit, S., Ritz, C. & Ramstein, G. Simulations of Northern Hemisphere ice-sheet retreat: oped in the periphery of the body14 during exhumation, but the Sensitivity to physical mechanisms involved during the Last Deglaciation. Quat. Sci. Rev. 21, 243–266 (2002). earlier olivine-rich microstructures and associated petrological 26. Crucifix, M., Loutre, M. F., Tulkens, P., Fichefet, T. & Berger, A. Climate evolution during the information are well preserved in the inner part (Figs 1, 2). Holocene: a study with an Earth system model of intermediate complexity. Clim. Dyn. 19, 43–60 Field observations and microstructural analyses allow two gen- (2002). 27. Crowley, T. Ice age terrestrial carbon changes revisited. Glob. Biogeochem. Cycles 9, 377–389 (1995). erations of tectonic fabrics to be distinguished in the dunite. The earliest D1 fabric is defined by the crystal-shape preferred orien- Acknowledgements We thank S. Hostetler for discussions and M. Siegert for comments and tation of coarse clear olivine grains (,0.6 mm) (Fig. 1a). A second suggestions. Model simulations were carried out at IDRIS/CNRS. This work was supported by the D2 fabric is widespread throughout the Higashi-akaishi body and ESF and the French national programmes ECLIPSE, PNEDC and ACI Jeunes Chercheurs. The consists of coarse dusty olivine porphyroclasts (,0.5 mm) and fine field work and other analyses were funded by the Research Council of Norway by grants to the project. M.J. was supported by NOAA. clear olivine neoblasts (,0.1 mm) (Fig. 1c), which formed by dynamic recrystallization of the D1 coarse-grained fabric (Fig. 1d). Competing interests statement The authors declare that they have no competing financial The preferred alignment of olivine crystals defines planar and linear interests. fabrics for both D1 (Fig. 1a) and D2 (Fig. 1c). The olivine grain Correspondence and requests for materials should be addressed to G.K. shape lineations are interpreted to be parallel to the maximum ([email protected]). extension direction. This interpretation is supported by the com- 432 © 2004 Nature Publishing Group NATURE | VOL 427 | 29 JANUARY 2004 | www.nature.com/nature